Recommendations for Starting with a Custom Industrial Log Dataset

[Tornadosky]

I'm about to start working with a custom real-world dataset. These are unstructured text-based machine logs collected from a car production plant. Before receiving the data, I would like to prepare a suitable pipeline for parsing and analyzing it (e.g., anomaly detection, failure prediction).

I’ve reviewed many papers and tools, including Drain, DeepLog, and LogLLM, but it’s still quite difficult to choose the most appropriate starting point for industrial logs.

Could you please advise on the following:

• Which existing datasets in Loghub are most suitable for initial testing, especially those that resemble industrial environments (e.g., high-frequency, noisy, complex systems)?

• What models or tools do you recommend for parsing and analyzing logs from a production plant?

• Are there any best practices or lessons learned when transitioning from academic datasets to custom industrial ones?